Hello,

I have just comited a new filter which has the same behavior as 
otbKeyPointSetsMatchingFilter, except this new one does not use brute force 
with L2 for descriptors matching.
This uses FLANN (included in OpenCV) to perform a nearest neighborhood 
approximation in the descriptor space. (see [Muja2009] 
<http://docs.opencv.org/modules/flann/doc/flann_fast_approximate_nearest_neighbor_search.html#id1>
 
Marius Muja, David G. Lowe. Fast Approximate Nearest Neighbors with 
Automatic Algorithm Configuration, 2009)
By the way, the code may need to be modified because I use it only with 
SIFT and I let the constant number of component at 128 for the descriptors 
size. No doubt it can work on other descriptors (SURF, ...)

The matching process is really, reeeeally faster than the classic 
otbKeyPointSetsMatchingFilter, but it also give more points and outliers 
(this seems to be inherent of the matching process, even with classic brute 
force in L2 same problem appears).
I noticed that the distance threshold in descriptor space has to be lower 
than for the classic otbKeyPointSetsMatchingFilter (it produces more 
matching points) 

For now, we use it in the framework of our project 
<http://geosud.teledetection.fr/> to perform dense and fast interest point 
extraction, and it seems to work nice.

Filter is in Code/FeatureExtraction near the classic one 
(otbKeyPointSetsFastMatchingFilter.h 
otbKeyPointSetsFastMatchingFilter.hxx). Its templated over the point set 
type only. BackMatching is implemented, and a threshold distance (in 
geographical domain) were added to perform distance-based selection in 
output.
I guess there is some work to integrate it properly. In addition, OpenCV is 
required. (I do not have included anything in the OTB cmake files, in order 
to not break anything :s)

We are really interested by some feedbacks and futures updates !

Thanks guys

Rémi



-- 
-- 
Check the OTB FAQ at
http://www.orfeo-toolbox.org/FAQ.html

You received this message because you are subscribed to the Google
Groups "otb-users" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/otb-users?hl=en
--- 
You received this message because you are subscribed to the Google Groups 
"otb-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to